| [P]arallel [Hi]gh-order [Li]brary for [P]DEs
    Latest
    Parallel High-Order Library for PDEs through hp-adaptive Discontinuous Galerkin methods | 
Negative Spalart-Allmaras model. Derived from Reynolds Averaged Navier Stokes. More...
#include <negative_spalart_allmaras_rans_model.h>
| Public Types | |
| using | thermal_boundary_condition_enum = Parameters::NavierStokesParam::ThermalBoundaryCondition | 
| using | two_point_num_flux_enum = Parameters::AllParameters::TwoPointNumericalFlux | 
|  Public Types inherited from PHiLiP::Physics::ReynoldsAveragedNavierStokesBase< dim, nstate, real > | |
| using | thermal_boundary_condition_enum = Parameters::NavierStokesParam::ThermalBoundaryCondition | 
| using | two_point_num_flux_enum = Parameters::AllParameters::TwoPointNumericalFlux | 
| Public Member Functions | |
| ReynoldsAveragedNavierStokes_SAneg (const Parameters::AllParameters *const parameters_input, const double ref_length, const double gamma_gas, const double mach_inf, const double angle_of_attack, const double side_slip_angle, const double prandtl_number, const double reynolds_number_inf, const bool use_constant_viscosity, const double constant_viscosity, const double turbulent_prandtl_number, const double temperature_inf=273.15, const double isothermal_wall_temperature=1.0, const thermal_boundary_condition_enum thermal_boundary_condition_type=thermal_boundary_condition_enum::adiabatic, std::shared_ptr< ManufacturedSolutionFunction< dim, real > > manufactured_solution_function=nullptr, const two_point_num_flux_enum two_point_num_flux_type=two_point_num_flux_enum::KG) | |
| dealii::Tensor< 2, dim, real > | compute_Reynolds_stress_tensor (const std::array< real, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, real >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< real, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized Reynolds stress tensor, (tau^reynolds)*, for the negative SA model. | |
| dealii::Tensor< 1, dim, real > | compute_Reynolds_heat_flux (const std::array< real, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, real >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< real, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized Reynolds heat flux, (q^reynolds)*, for the negative SA model. | |
| dealii::Tensor< 2, dim, FadType > | compute_Reynolds_stress_tensor_fad (const std::array< FadType, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, FadType >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< FadType, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized Reynolds stress tensor, (tau^reynolds)* (Automatic Differentiation Type: FadType), for the negative SA model. | |
| dealii::Tensor< 1, dim, FadType > | compute_Reynolds_heat_flux_fad (const std::array< FadType, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, FadType >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< FadType, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized Reynolds heat flux, (q^reynolds)* (Automatic Differentiation Type: FadType), for the negative SA model. | |
| real | compute_eddy_viscosity (const std::array< real, nstate_navier_stokes > &primitive_soln_rans, const std::array< real, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized eddy viscosity for the negative SA model. | |
| FadType | compute_eddy_viscosity_fad (const std::array< FadType, nstate_navier_stokes > &primitive_soln_rans, const std::array< FadType, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized eddy viscosity for the negative SA model (Automatic Differentiation Type: FadType) | |
| std::array< real, nstate-(dim+2)> | compute_effective_viscosity_turbulence_model (const std::array< real, nstate_navier_stokes > &primitive_soln_rans, const std::array< real, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized effective (total) viscosities for the negative SA model. | |
| std::array< FadType, nstate-(dim+2)> | compute_effective_viscosity_turbulence_model_fad (const std::array< FadType, nstate_navier_stokes > &primitive_soln_rans, const std::array< FadType, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Nondimensionalized effective (total) viscosities for the negative SA model (Automatic Differentiation Type: FadType) | |
| std::array< real, nstate > | compute_production_dissipation_cross_term (const dealii::Point< dim, real > &pos, const std::array< real, nstate > &conservative_solution, const std::array< dealii::Tensor< 1, dim, real >, nstate > &solution_gradient) const | 
| Physical source term (production, dissipation source terms and source term with cross derivatives) for the negative SA model. | |
| dealii::Vector< double > | post_compute_derived_quantities_vector (const dealii::Vector< double > &uh, const std::vector< dealii::Tensor< 1, dim > > &, const std::vector< dealii::Tensor< 2, dim > > &, const dealii::Tensor< 1, dim > &, const dealii::Point< dim > &) const override | 
| For post processing purposes, returns conservative and primitive solution variables for the negative SA model. | |
| std::vector< dealii::DataComponentInterpretation::DataComponentInterpretation > | post_get_data_component_interpretation () const override | 
| For post processing purposes, sets the interpretation of each computed quantity as either scalar or vector for the negative SA model. | |
| std::vector< std::string > | post_get_names () const override | 
| For post processing purposes, sets the base names (with no prefix or suffix) of the computed quantities for the negative SA model. | |
|  Public Member Functions inherited from PHiLiP::Physics::ReynoldsAveragedNavierStokesBase< dim, nstate, real > | |
| ReynoldsAveragedNavierStokesBase (const Parameters::AllParameters *const parameters_input, const double ref_length, const double gamma_gas, const double mach_inf, const double angle_of_attack, const double side_slip_angle, const double prandtl_number, const double reynolds_number_inf, const bool use_constant_viscosity, const double constant_viscosity, const double turbulent_prandtl_number, const double temperature_inf=273.15, const double isothermal_wall_temperature=1.0, const thermal_boundary_condition_enum thermal_boundary_condition_type=thermal_boundary_condition_enum::adiabatic, std::shared_ptr< ManufacturedSolutionFunction< dim, real > > manufactured_solution_function=nullptr, const two_point_num_flux_enum two_point_num_flux_type=two_point_num_flux_enum::KG) | |
| Constructor. | |
| std::array< dealii::Tensor< 1, dim, real >, nstate > | convective_flux (const std::array< real, nstate > &conservative_soln) const | 
| Additional convective flux of RANS + convective flux of turbulence model. | |
| std::array< dealii::Tensor< 1, dim, real >, nstate > | dissipative_flux (const std::array< real, nstate > &conservative_soln, const std::array< dealii::Tensor< 1, dim, real >, nstate > &solution_gradient, const dealii::types::global_dof_index cell_index) const | 
| Additional viscous flux of RANS + viscous flux of turbulence model. | |
| std::array< real, nstate > | convective_eigenvalues (const std::array< real, nstate > &, const dealii::Tensor< 1, dim, real > &) const | 
| Convective eigenvalues of the additional models' PDEs.  More... | |
| real | max_convective_eigenvalue (const std::array< real, nstate > &soln) const | 
| Maximum convective eigenvalue of the additional models' PDEs.  More... | |
| real | max_convective_normal_eigenvalue (const std::array< real, nstate > &soln, const dealii::Tensor< 1, dim, real > &normal) const | 
| Maximum convective normal eigenvalue (used in Lax-Friedrichs) of the additional models' PDEs.  More... | |
| std::array< real, nstate > | source_term (const dealii::Point< dim, real > &pos, const std::array< real, nstate > &conservative_solution, const real current_time, const dealii::types::global_dof_index cell_index) const | 
| Source term for manufactured solution functions. | |
| std::array< real, nstate > | convective_dissipative_source_term (const dealii::Point< dim, real > &pos, const std::array< real, nstate > &conservative_solution, const dealii::types::global_dof_index cell_index) const | 
| Convective and dissipative source term for manufactured solution functions. | |
| std::array< real, nstate > | physical_source_term (const dealii::Point< dim, real > &pos, const std::array< real, nstate > &conservative_solution, const std::array< dealii::Tensor< 1, dim, real >, nstate > &solution_gradient, const dealii::types::global_dof_index cell_index) const override | 
| Physical source term. | |
|  Public Member Functions inherited from PHiLiP::Physics::ModelBase< dim, nstate, real > | |
| ModelBase (std::shared_ptr< ManufacturedSolutionFunction< dim, real > > manufactured_solution_function_input=nullptr) | |
| Constructor. | |
| virtual | ~ModelBase ()=default | 
| Virtual destructor required for abstract classes. | |
| void | boundary_face_values (const int boundary_type, const dealii::Point< dim, real > &pos, const dealii::Tensor< 1, dim, real > &normal, const std::array< real, nstate > &soln_int, const std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_int, std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const | 
| Boundary condition handler. | |
| Static Public Attributes | |
| static const int | nstate_navier_stokes = dim+2 | 
| Number of PDEs for RANS equations. | |
| static const int | nstate_turbulence_model = nstate-(dim+2) | 
| Number of PDEs for RANS turbulence model. | |
|  Static Public Attributes inherited from PHiLiP::Physics::ReynoldsAveragedNavierStokesBase< dim, nstate, real > | |
| static const int | nstate_navier_stokes = dim+2 | 
| Number of PDEs for RANS equations. | |
| static const int | nstate_turbulence_model = nstate-(dim+2) | 
| Number of PDEs for RANS turbulence model. | |
| Protected Member Functions | |
| template<typename real2 > | |
| dealii::Tensor< 2, dim, real2 > | compute_Reynolds_stress_tensor_templated (const std::array< real2, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, real2 >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< real2, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Templated nondimensionalized Reynolds stress tensor, (tau^reynolds)* for the negative SA model. | |
| template<typename real2 > | |
| dealii::Tensor< 1, dim, real2 > | compute_Reynolds_heat_flux_templated (const std::array< real2, nstate_navier_stokes > &primitive_soln_rans, const std::array< dealii::Tensor< 1, dim, real2 >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< real2, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Templated nondimensionalized Reynolds heat flux, (q^reynolds)* for the negative SA model. | |
| template<typename real2 > | |
| real2 | scale_coefficient (const real2 coefficient) const | 
| Templated nondimensionalized variables scaled by reynolds_number_inf for the negative SA model. | |
| void | boundary_wall (std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const override | 
| Wall boundary condition.  More... | |
| void | boundary_outflow (const std::array< real, nstate > &soln_int, const std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_int, std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const override | 
| Outflow boundary condition. | |
| void | boundary_inflow (const std::array< real, nstate > &soln_int, const std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_int, std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const override | 
| Inflow boundary condition.  More... | |
| void | boundary_farfield (std::array< real, nstate > &soln_bc) const override | 
| Farfield boundary conditions based on freestream values.  More... | |
|  Protected Member Functions inherited from PHiLiP::Physics::ReynoldsAveragedNavierStokesBase< dim, nstate, real > | |
| template<typename real2 > | |
| real2 | get_vector_magnitude_sqr (const dealii::Tensor< 1, 3, real2 > &vector) const | 
| Returns the square of the magnitude of the vector. | |
| template<typename real2 > | |
| real2 | get_tensor_magnitude_sqr (const dealii::Tensor< 2, dim, real2 > &tensor) const | 
| Returns the square of the magnitude of the tensor (i.e. the double dot product of a tensor with itself) | |
| template<typename real2 > | |
| std::array< dealii::Tensor< 1, dim, real2 >, nstate > | dissipative_flux_templated (const std::array< real2, nstate > &conservative_soln, const std::array< dealii::Tensor< 1, dim, real2 >, nstate > &solution_gradient, const dealii::types::global_dof_index cell_index) const | 
| Templated additional dissipative (i.e. viscous) flux. | |
| template<typename real2 > | |
| std::array< dealii::Tensor< 1, dim, real2 >, nstate > | convective_flux_templated (const std::array< real2, nstate > &conservative_soln) const | 
| Templated additional convective flux. | |
| template<typename real2 > | |
| std::array< real2, dim+2 > | extract_rans_conservative_solution (const std::array< real2, nstate > &conservative_soln) const | 
| Returns the conservative solutions of Reynolds-averaged Navier-Stokes equations (without additional RANS turbulence model) | |
| template<typename real2 > | |
| std::array< dealii::Tensor< 1, dim, real2 >, dim+2 > | extract_rans_solution_gradient (const std::array< dealii::Tensor< 1, dim, real2 >, nstate > &solution_gradient) const | 
| Returns the conservative solutions gradient of Reynolds-averaged Navier-Stokes equations (without additional RANS turbulence model) | |
| template<typename real2 > | |
| std::array< dealii::Tensor< 1, dim, real2 >, nstate-(dim+2)> | dissipative_flux_turbulence_model (const std::array< real2, nstate_navier_stokes > &primitive_soln_rans, const std::array< real2, nstate_turbulence_model > &primitive_soln_turbulence_model, const std::array< dealii::Tensor< 1, dim, real2 >, nstate_turbulence_model > &primitive_solution_gradient_turbulence_model) const | 
| Templated Additional viscous flux of RANS + viscous flux of turbulence model. | |
| template<typename real2 > | |
| std::array< real2, nstate-(dim+2)> | convert_conservative_to_primitive_turbulence_model (const std::array< real2, nstate > &conservative_soln) const | 
| template<typename real2 > | |
| std::array< dealii::Tensor< 1, dim, real2 >, nstate-(dim+2)> | convert_conservative_gradient_to_primitive_gradient_turbulence_model (const std::array< real2, nstate > &conservative_soln, const std::array< dealii::Tensor< 1, dim, real2 >, nstate > &solution_gradient) const | 
| std::array< real, nstate-(dim+2)> | compute_mean_turbulence_property (const std::array< real, nstate > &conservative_soln1, const std::array< real, nstate > &conservative_soln2) const | 
| Mean turbulence properties given two sets of conservative solutions.  More... | |
| dealii::Tensor< 2, nstate, real > | convective_flux_directional_jacobian (const std::array< real, nstate > &conservative_soln, const dealii::Tensor< 1, dim, real > &normal) const | 
| dealii::Tensor< 2, nstate, real > | dissipative_flux_directional_jacobian (const std::array< real, nstate > &conservative_soln, const std::array< dealii::Tensor< 1, dim, real >, nstate > &solution_gradient, const dealii::Tensor< 1, dim, real > &normal, const dealii::types::global_dof_index cell_index) const | 
| dealii::Tensor< 2, nstate, real > | dissipative_flux_directional_jacobian_wrt_gradient_component (const std::array< real, nstate > &conservative_soln, const std::array< dealii::Tensor< 1, dim, real >, nstate > &solution_gradient, const dealii::Tensor< 1, dim, real > &normal, const int d_gradient, const dealii::types::global_dof_index cell_index) const | 
| std::array< real, nstate > | get_manufactured_solution_value (const dealii::Point< dim, real > &pos) const | 
| Get manufactured solution value. | |
| std::array< dealii::Tensor< 1, dim, real >, nstate > | get_manufactured_solution_gradient (const dealii::Point< dim, real > &pos) const | 
| Get manufactured solution value. | |
| std::array< real, nstate > | convective_source_term_computed_from_manufactured_solution (const dealii::Point< dim, real > &pos) const | 
| std::array< real, nstate > | dissipative_source_term_computed_from_manufactured_solution (const dealii::Point< dim, real > &pos, const dealii::types::global_dof_index cell_index) const | 
| std::array< real, nstate > | physical_source_term_computed_from_manufactured_solution (const dealii::Point< dim, real > &pos, const dealii::types::global_dof_index cell_index) const | 
| void | boundary_manufactured_solution (const dealii::Point< dim, real > &pos, const dealii::Tensor< 1, dim, real > &normal_int, const std::array< real, nstate > &soln_int, const std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_int, std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const override | 
| Evaluate the manufactured solution boundary conditions. | |
|  Protected Member Functions inherited from PHiLiP::Physics::ModelBase< dim, nstate, real > | |
| virtual void | boundary_riemann (const dealii::Tensor< 1, dim, real > &normal_int, const std::array< real, nstate > &soln_int, std::array< real, nstate > &soln_bc) const | 
| virtual void | boundary_slip_wall (const dealii::Tensor< 1, dim, real > &normal_int, const std::array< real, nstate > &soln_int, const std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_int, std::array< real, nstate > &soln_bc, std::array< dealii::Tensor< 1, dim, real >, nstate > &soln_grad_bc) const | 
| Private Member Functions | |
| template<typename real2 > | |
| real2 | compute_eddy_viscosity_templated (const std::array< real2, nstate_navier_stokes > &primitive_soln_rans, const std::array< real2, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Templated nondimensionalized eddy viscosity for the negative SA model.  More... | |
| template<typename real2 > | |
| std::array< real2, nstate-(dim+2)> | compute_effective_viscosity_turbulence_model_templated (const std::array< real2, nstate_navier_stokes > &primitive_soln_rans, const std::array< real2, nstate_turbulence_model > &primitive_soln_turbulence_model) const | 
| Templated nondimensionalized effective (total) viscosities for the negative SA model.  More... | |
| template<typename real2 > | |
| real2 | compute_coefficient_Chi (const real2 nu_tilde, const real2 laminar_kinematic_viscosity) const | 
| Templated coefficient Chi for the negative SA model.  More... | |
| template<typename real2 > | |
| real2 | compute_coefficient_f_v1 (const real2 coefficient_Chi) const | 
| Templated coefficient f_v1 for the negative SA model.  More... | |
| real | compute_coefficient_f_v2 (const real coefficient_Chi) const | 
| Coefficient f_v2 for the negative SA model.  More... | |
| template<typename real2 > | |
| real2 | compute_coefficient_f_n (const real2 nu_tilde, const real2 laminar_kinematic_viscosity) const | 
| Templated coefficient f_n for the negative SA model.  More... | |
| real | compute_coefficient_f_t2 (const real coefficient_Chi) const | 
| Coefficient f_t2 for the negative SA model.  More... | |
| real | compute_coefficient_f_w (const real coefficient_g) const | 
| Coefficient f_t2 for the negative SA model.  More... | |
| real | compute_coefficient_r (const real nu_tilde, const real d_wall, const real s_tilde) const | 
| Coefficient r for the negative SA model.  More... | |
| real | compute_coefficient_g (const real coefficient_r) const | 
| Coefficient g for the negative SA model.  More... | |
| real | compute_s (const std::array< real, nstate_navier_stokes > &conservative_soln_rans, const std::array< dealii::Tensor< 1, dim, real >, nstate_navier_stokes > &conservative_soln_gradient_rans) const | 
| Vorticity magnitude for the negative SA model, sqrt(vorticity_x^2+vorticity_y^2+vorticity_z^2) | |
| real | compute_s_bar (const real coefficient_Chi, const real nu_tilde, const real d_wall) const | 
| Correction of vorticity magnitude for the negative SA model.  More... | |
| real | compute_s_tilde (const real coefficient_Chi, const real nu_tilde, const real d_wall, const real s) const | 
| Modified vorticity magnitude for the negative SA model.  More... | |
| std::array< real, nstate > | compute_production_source (const real coefficient_f_t2, const real density, const real nu_tilde, const real s, const real s_tilde) const | 
| Production source term for the negative SA model.  More... | |
| std::array< real, nstate > | compute_dissipation_source (const real coefficient_f_t2, const real density, const real nu_tilde, const real d_wall, const real s_tilde) const | 
| Dissipation source term for the negative SA model.  More... | |
| std::array< real, nstate > | compute_cross_source (const real density, const real nu_tilde, const real laminar_kinematic_viscosity, const std::array< dealii::Tensor< 1, dim, real >, nstate_navier_stokes > &primitive_soln_gradient_rans, const std::array< dealii::Tensor< 1, dim, real >, nstate_turbulence_model > &primitive_solution_gradient_turbulence_model) const | 
| Source term with cross derivatives for the negative SA model.  More... | |
| Private Attributes | |
| const real | c_b1 = 0.1355 | 
| Constant coefficients for the negative SA model.  More... | |
| const real | sigma = 2.0/3.0 | 
| const real | c_b2 = 0.622 | 
| const real | kappa = 0.41 | 
| const real | c_w1 = c_b1/(kappa*kappa)+(1+c_b2)/sigma | 
| const real | c_w2 = 0.3 | 
| const real | c_w3 = 2.0 | 
| const real | c_v1 = 7.1 | 
| const real | c_v2 = 0.7 | 
| const real | c_v3 = 0.9 | 
| const real | c_t3 = 1.2 | 
| const real | c_t4 = 0.5 | 
| const real | c_n1 = 16.0 | 
| const real | r_lim = 10.0 | 
| const FadType | sigma_fad = 2.0/3.0 | 
| const FadType | c_v1_fad = 7.1 | 
| const FadType | c_n1_fad = 16.0 | 
| Additional Inherited Members | |
|  Public Attributes inherited from PHiLiP::Physics::ReynoldsAveragedNavierStokesBase< dim, nstate, real > | |
| const double | turbulent_prandtl_number | 
| Turbulent Prandtl number. | |
| std::unique_ptr< NavierStokes< dim, nstate_navier_stokes, real > > | navier_stokes_physics | 
| Pointer to Navier-Stokes physics object. | |
|  Public Attributes inherited from PHiLiP::Physics::ModelBase< dim, nstate, real > | |
| std::shared_ptr< ManufacturedSolutionFunction< dim, real > > | manufactured_solution_function | 
| Manufactured solution function. | |
| dealii::LinearAlgebra::distributed::Vector< int > | cellwise_poly_degree | 
| Cellwise polynomial degree. | |
| dealii::LinearAlgebra::distributed::Vector< double > | cellwise_volume | 
|  Protected Attributes inherited from PHiLiP::Physics::ModelBase< dim, nstate, real > | |
| const MPI_Comm | mpi_communicator | 
| MPI communicator. | |
| dealii::ConditionalOStream | pcout | 
| Parallel std::cout that only outputs on mpi_rank==0. | |
Negative Spalart-Allmaras model. Derived from Reynolds Averaged Navier Stokes.
Definition at line 14 of file negative_spalart_allmaras_rans_model.h.
| PHiLiP::Physics::ReynoldsAveragedNavierStokes_SAneg< dim, nstate, real >::ReynoldsAveragedNavierStokes_SAneg | ( | const Parameters::AllParameters *const | parameters_input, | 
| const double | ref_length, | ||
| const double | gamma_gas, | ||
| const double | mach_inf, | ||
| const double | angle_of_attack, | ||
| const double | side_slip_angle, | ||
| const double | prandtl_number, | ||
| const double | reynolds_number_inf, | ||
| const bool | use_constant_viscosity, | ||
| const double | constant_viscosity, | ||
| const double | turbulent_prandtl_number, | ||
| const double | temperature_inf = 273.15, | ||
| const double | isothermal_wall_temperature = 1.0, | ||
| const thermal_boundary_condition_enum | thermal_boundary_condition_type = thermal_boundary_condition_enum::adiabatic, | ||
| std::shared_ptr< ManufacturedSolutionFunction< dim, real > > | manufactured_solution_function = nullptr, | ||
| const two_point_num_flux_enum | two_point_num_flux_type = two_point_num_flux_enum::KG | ||
| ) | 
Constructor for the Reynolds-averaged Navier-Stokes model: negative SA Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model."
Definition at line 18 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | overrideprotectedvirtual | 
Farfield boundary conditions based on freestream values.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(8)
Reimplemented from PHiLiP::Physics::ModelBase< dim, nstate, real >.
Definition at line 646 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | overrideprotectedvirtual | 
Inflow boundary condition.
Reference: Philippe R. Spalart. (2007). "Effective Inflow Conditions for Turbulence Models in Aerodynamc Calculations." Section III. Ambient Values: A. Turbulent Reynolds Numbers
Reimplemented from PHiLiP::Physics::ModelBase< dim, nstate, real >.
Definition at line 629 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | overrideprotectedvirtual | 
Wall boundary condition.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(7)
Reimplemented from PHiLiP::Physics::ModelBase< dim, nstate, real >.
Definition at line 601 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Templated coefficient Chi for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(1)
Definition at line 168 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Templated coefficient f_n for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(21)
Definition at line 219 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Coefficient f_t2 for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(6)
Definition at line 250 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Templated coefficient f_v1 for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(1)
Definition at line 181 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Coefficient f_v2 for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(4)
Definition at line 204 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Coefficient f_t2 for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(5)
Definition at line 263 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Coefficient g for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(5)
Definition at line 293 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Coefficient r for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(5)
Definition at line 276 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Source term with cross derivatives for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(17) for positive nu_tilde
Definition at line 418 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Dissipation source term for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(3) for positive nu_tilde eq.(22) for negative nu_tilde
Definition at line 389 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Templated nondimensionalized eddy viscosity for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(1)
Definition at line 74 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Templated nondimensionalized effective (total) viscosities for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(14)
Definition at line 137 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Production source term for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(3) for positive nu_tilde eq.(22) for negative nu_tilde
Definition at line 364 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Correction of vorticity magnitude for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(11)
Definition at line 324 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Modified vorticity magnitude for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model." eq.(12)
Definition at line 340 of file negative_spalart_allmaras_rans_model.cpp.
| 
 | private | 
Constant coefficients for the negative SA model.
Reference: Steven R. Allmaras. (2012). "Modifications and Clarifications for the Implementation of the Spalart-Allmaras Turbulence Model."
Definition at line 298 of file negative_spalart_allmaras_rans_model.h.